Running recommendation models on AMD CPUs

[rishucoding]

Hi,

I am interested in running Recommendation models like Deep Learning Recommendation Models (DLRM) on AMD CPUs. Like, I would like to run MLPerf DLRM or DeepRecSys.
I am wondering if AMD provides specific guidelines/support (CPU-tuning, libraries, etc) similar to Intel Zoo for optimizing inference performance?

Thanks,
Rishabh Jain

[varunsh-xilinx]

I've asked the ZenDNN team to comment. In the meantime, you can check the ZenDNN documentation if you haven't already.

[ratan-prasad]

Since DLRM is a shallow model, we need to put multiple instances simultaneously to get optimal performance. We are in the process of creating scripts for multi-instance runs and tuning guide for optimal performance. Hopefully, this will be released with the next ZenDNN release.

[rishucoding]

Thanks @ratan-prasad. I have a few follow-up questions:

1. What does a shallow model mean? Does it mean small in size?
2. For putting multiple instances on a same CPU core, are you suggesting to explore various ways to deploy multi-threading?